Researchers at West Virginia University report that an autonomous registry powered by generative Artificial Intelligence can ease the burden of clinical decision-making in drug-resistant epilepsy by extracting and organizing key data from fragmented medical records. Presented at the American Epilepsy Society annual meeting, the pilot study evaluated the Autonomous Registry and Analytics platform, or AURA, which is integrated into the EPIC electronic health record and continuously analyzes free-text clinical notes, imaging reports and assessments for patients with drug-resistant epilepsy. P. David Adelson, MD, who led the work, said epilepsy care generates an extraordinary volume of nuanced information that is carefully documented but difficult to track across years of notes during routine clinical practice, and that even expert teams can miss opportunities when crucial signals are buried in narrative documentation.
The AURA registry currently includes 3,348 patients (pediatric, n = 1,049; adult, n = 2,049), and the platform recognizes indicators such as multiple failed antiseizure medications, seizure burden and diagnostic results that may suggest suitability for surgery. For incomplete or unclear records, such as missing MRI, EEG or neuropsychological evaluations, the system queries prior encounters, extracts relevant data and flags unresolved gaps, and it can also detect unmet standard-of-care interventions like folic acid supplementation for women of reproductive age. The primary outcome of the pilot, which analyzed documentation of 820 scheduled visits, was to assess AURA’s impact by tracking care gaps and pre-visit readiness, along with clinician-reported usefulness for neurosurgical decision-making, within an “always-on” support layer that surfaces what is already in the chart when decisions are required.
Over the 90-day trial, AURA identified 88 patients (11%) whose medical record exhibited clinical patterns consistent with drug-resistant epilepsy but lacked the requisite surgical referrals. AURA flagged the following percentages of EHR-related care gaps related to diagnostic workups: missing EEGs (35%), outdated or missing MRIs (54%), absent neuropsychological evaluations (91%), and folic acid deficiencies (81%). The researchers found that clinicians using AURA reported improved pre-visit readiness, reduced chart review time and more timely identification of surgical candidates, and clinicians’ post-surgical outcome documentation increased from 1% to 70.9%. Adelson said these large language model agents can meaningfully augment, not replace, epilepsy specialists by functioning like an army of highly trained fellows that continuously read charts and flag issues at the right moment, reducing cognitive burden and variability while supporting faster, safer and more coordinated care across epilepsy programs.
